BP神经网络在雌二醇衍生物QSAR中的应用研究

Study on the QSAR of Estradiol Derivatives Using the Backpropagation Neural Network

目的:探讨误差反向传播(backpropagation,BP)神经网络在雌二醇衍生物定量结构-活性关系(quantitative structure-activity relationships,QSAR)研究中的应用。方法:采用BP神经网络法和多元线性回归法,分别建立了30个雌二醇衍生物在0℃下与羔羊子宫雌激素受体间亲合力logRBA与疏水性参数logP、分子的体积V和9号碳原子的净电荷Q之间的QSAR模型。结果:BP模型的相关系数R=0.9962,标准偏差SD=0.0588;MLR模型的相关系数R=0.9090,标准偏差SD=0.2904。结论:BP神经网络是一种比较精密的拟合方法,具有良好的预测效果。

Objective,Explore BP（backpropagation） the nerve network in the estradiol derivatives QSAR（quantitative structure-activity relationships） research applications. Methods：The quantitative structure-activity relationship （QSAR） of 30 estradiol derivatives was studied with BP and MLR method. Three parameters as input were hydrophobic parameter （logP） ,net charge on the 9th carbon （Q） and the volume of molecule （V）. The affinity of compound at 0 ℃ acting on estrogen acceptor in calf uterine tissue was used as biological activity （logRBA）. Based on the improved back-propagation （BP） algorithm, the QSAR model was set up. Results：It was obtained that the correlation coefficient was R= 0. 9962 and the standard deviation was SD= 0. 0588. For 30 estradiol derivatives, linear regression equation was obtained meanwhile. Its correlation coefficient R is equal to 0, 9090 and the standard error SD to 0, 2904. Conclusion： The result showed that the fitted performance of BP network method was comparatively precise and it has a preferable predicted effect.